Hanish's Garden

🌿 It's a wave - doing anything with audio

Properties1
tagscheatsheet

3 min read

The command-line moves I reach for when I want to wrangle audio outside a DAW - quick conversions, cleanups, and measurements. Two tools cover almost everything:

  • FFmpeg - formats, containers, loudness, pulling audio out of video. (For the video side, see the FFmpeg guide.)
  • SoX - “Sound eXchange,” the DSP swiss-army knife: effects, silence trimming, noise reduction, analysis. apt install sox.

Rule of thumb: FFmpeg to convert/normalize/extract, SoX for effects and signal processing.

Convert, resample, re-channel

ffmpeg -i in.wav -c:a libmp3lame -q:a 2 out.mp3     # WAV → MP3 (q:a 0 best … 9 worst)
ffmpeg -i in.wav -c:a flac out.flac                 # lossless FLAC
ffmpeg -i in.mp3 -ar 44100 -ac 2 out.wav            # resample to 44.1 kHz, force stereo
sox in.wav -c 1 mono.wav                            # downmix to mono

Trim, join, split

sox in.wav clip.wav trim 10 30                      # 30s starting at 0:10
sox a.wav b.wav c.wav joined.wav                    # concatenate (same format)
# Split a long file at every silence gap into separate files:
sox in.wav out.wav silence 1 0.5 1% 1 0.5 1% : newfile : restart

FFmpeg equivalents for cutting/joining live in the FFmpeg guide.

Volume & loudness

sox in.wav out.wav gain -n                          # peak-normalize to 0 dBFS
sox in.wav out.wav gain -3                          # attenuate by 3 dB
ffmpeg -i in.wav -af "volume=1.5" louder.wav        # ×1.5 gain
 
# Broadcast/podcast loudness normalize (EBU R128) - the right way to "make it consistent"
ffmpeg -i in.wav -af loudnorm=I=-16:TP=-1.5:LRA=11 out.wav

Peak vs loudness

gain -n maxes the peak - but two files normalized that way can still feel like very different volumes. loudnorm targets perceived loudness (LUFS), which is what streaming/podcast platforms actually judge you on (−14 to −16 LUFS).

Fades & silence

sox in.wav out.wav fade t 3 0 3                      # 3s fade-in, 3s fade-out
sox in.wav out.wav silence 1 0.1 1% reverse silence 1 0.1 1% reverse   # trim lead/trail silence
ffmpeg -i in.wav -af silencedetect=n=-30dB:d=0.5 -f null -            # find silent spots (prints timestamps)

Clean it up - noise reduction (SoX, two passes)

# 1. Profile the noise from a silent/room-tone section (first 0.5s here)
sox in.wav -n trim 0 0.5 noiseprof noise.prof
# 2. Subtract it (0.21 = strength; go gentle or you get underwater artifacts)
sox in.wav clean.wav noisered noise.prof 0.21

Pitch & tempo (independently)

sox in.wav out.wav pitch 200                        # +200 cents (2 semitones), tempo unchanged
sox in.wav out.wav tempo 1.25                        # 25% faster, pitch unchanged
# FFmpeg tempo (0.5–2.0 per stage; chain for more):
ffmpeg -i in.wav -af "atempo=1.5" fast.wav

Measure & inspect

sox in.wav -n stat                                   # peak, RMS, duration, DC offset…
sox in.wav -n stats                                  # dBFS peak/RMS, crest factor
ffmpeg -i in.wav -af volumedetect -f null -          # mean/max volume (great before normalizing)
ffprobe -hide_banner in.wav                          # codec, sample rate, channels, bitrate

Generate & metadata

sox -n tone.wav synth 3 sine 440                     # 3s 440 Hz sine (test tones, clicks, silence)
ffmpeg -i in.mp3 -metadata title="Take 3" -metadata artist="Me" -c copy tagged.mp3

Notes

  • SoX effects chain left→right - sox in out gain -3 fade t 2 0 2 pitch 100 applies gain, then fade, then pitch, in that order. Order matters.
  • Work in WAV/FLAC while editing, export to MP3/AAC last - every lossy re-encode throws away quality you can’t get back.
  • For anything visual (video audio tracks, muxing, subtitles) jump to the FFmpeg guide.

Graph View

Backlinks